App performance on Pentium M 760 vs. Pentium 4 3.6GHz processor

App performance on Pentium M 760 vs. Pentium 4 3.6GHz processor

The following is a question recently received by Intel Software Network Support, followed by the response providedby our engineering team:

==

Q. We have an application that runs a process in 85 seconds on a Pentium M 760 processor and takes 100 seconds on a Pentium 4 3.6GHz processor-based desktop with a faster hard drive. Is there anything different about the PentiumM processors that accounts for their speed difference in spite of a lower clock speed? The Pentium 4 processor has 2MB L2 cache, the same as the Pentium M 760 processor.

A. Many factors besides gigahertz determine the relative performance of a microprocessor.

To give an analogy, a microprocessor is like a highway, only you dont know when to turn until you are at the turn. You can try to guess which turn to take and if you guess right you stay at speed. If you guess wrong and miss the turn, you have to stop and go back to the turn and get up to speed again.

The Intel Pentium 4 processor goes faster, but it has big penalties for missing the turn. It also takes much more energy to accelerate to speed (again, think about why you get better mileage on the highway versus city). That energy also creates a lot of heat, so the Intel Pentium 4 creates a lot of heat and uses a lot of watts.

The Intel Pentium M processor was designed for mobiles where heat creation and watt usage are bad, so it has a slower top speed. It compensates by improving the branch predictor (so it misses fewer turns) and having better acceleration (by reducing latency for loading instructions through a different cache architecture).

The new Intel Core 2 Architecture combines techniques from both processors, with a top speed almost as fast the Intel Pentium 4 processor and the improved branch predictor and cache of the Intel Pentium M processor. It adds one more feature to improve performance - a wider instruction bus. All current microprocessors can handle three instructions per clock (to push my analogy, it is like a three lane highway), whereas the Intel Core 2 Duo processor handles four instructions per clock (like adding a lane to the highway, more work can be handled in the same amount of time). The Intel Core 2 Architecture is available now as the Intel Xeon 5100 series in servers and will soon be available as the Intel Core 2 Duo in desktops and notebooks.

Finally, all the new processors are dual-core. This gives you an extra processor on the same piece of silicon to handle additional work.

In all, it is not surprising that gigahertz did not correlate directly with performance, and we expect the correlation to decrease even more over time as new techniques besides raw speed are used to improve application performance.

===

Lexi S.

IntelSoftware NetworkSupport

http://www.intel.com/software

Contact us

1 post / 0 new
For more complete information about compiler optimizations, see our Optimization Notice.